About the detect dataset in pretrain stage

[zhww]

In the technical report，Object365 was used for pretrain, but I test mini-4B, it can't do object detection. So, what the detect data format or instructions are used during pretrain?

[czczup]

InternVL 1.5 does not support object detection because we only used object detection data during the pre-training stage and did not utilize it during the SFT stage.

Now, the new InternVL 2.0 series models support object detection. You can try it out in our online demo: https://internvl.opengvlab.com/